Computer-aided detection software does not help clinicians analyze and interpret film-screen screening mammograms, results from a large multicenter study demonstrated.
The findings "raise concerns that CAD, as currently implemented in clinical practice, may have little or no impact on breast cancer mortality, which may depend on earlier detection of invasive breast cancer," researchers led by Dr. Joshua J. Fenton reported online July 27 in the Journal of the National Cancer Institute.
Dr. Fenton of the department of family and community medicine at the center for health care policy and research of the University of California, Davis, and his associates analyzed records from 684,956 women who underwent more than 1.6 million film-screen screening mammograms in 1998-2006 at 90 facilities that are members of the Breast Cancer Surveillance Consortium (BCSC), a federally supported effort that links mammography data to cancer outcomes in seven states (California, Colorado, North Carolina, New Hampshire, New Mexico, Washington, and Vermont).
The researchers used random effects logistic regression to estimate the associations between CAD and specificity, sensitivity, and positive predictive value while adjusting for factors that might influence mammography findings, including patient age, breast density, and use of hormone therapy (J. Natl. Cancer Inst. 2011 July 27 [doi:10.1093/jnci/djr298]).
More than half of the women studied (61%) were aged 40-59 years. Of the 90 BCSC facilities, 25 (28%) implemented CAD and used it for an average of 28 months during the study period. All told, 793 radiologists interpreted the results, including 154 (19%) at facilities with CAD.
After adjusting for BCSC registry, patient characteristics, hormone therapy use, and year of mammography interpretation, the researchers found that CAD use was associated with a statistically significant lower specificity (odds ratio, 0.87) and positive predictive value (OR, 0.89), and an increase in sensitivity (OR, 1.06) that was not statistically significant.
When the sensitivity analysis was limited to invasive cancers only, CAD use was no longer associated with increased sensitivity (OR, 0.96). When the analysis was limited to ductal carcinoma in situ, CAD use was associated with greater sensitivity (OR, 1.55), yet this did not reach statistical significance.
There were also no statistically significant differences between CAD use and no-CAD in the odds of overall breast cancer detection (OR, 1.01), the diagnosis of stage I invasive cancer compared with later-stage invasive cancer (OR, 0.97), or the diagnosis of invasive tumors of 15 mm or less in size, compared with those greater than 15 mm (OR, 0.92).
Dr. Fenton and his associates acknowledged certain limitations of the study, including the absence of digital mammography data. "Whereas CAD algorithms perform a similar alerting function in the film-screen and digital environments, film-screen mammograms must be digitized before CAD analysis, and digitization may introduce noise and adversely affect performance," they wrote.
"However, small retrospective studies suggest that the performance impacts of CAD are similar when used in digital and film-screen environments. Because prior research suggests that facilities apply CAD on nearly all mammograms after implementation, these analyses assumed that all mammograms were interpreted with CAD after implementation – another limitation of this study."
The study was supported by grants from the National Cancer Institute and the American Cancer Society. A study coauthor disclosed holding stock options and serving as a medical consultant for Hologic Inc., a manufacturer of CAD equipment that had "no role in data collection, analysis, or interpretation."